Enhanced Fatigue Life Prediction via Spatiotemporal Convolutional Autoencoder-Driven Crack Growth Modeling

**Abstract:** This study introduces a novel methodology for predicting fatigue life and crack growth behavior in metallic materials, specifically focusing on high-cycle fatigue (HCF) scenarios. Leveraging a Spatiotemporal Convolutional Autoencoder (STCAE) architecture trained on high-resolution digital image correlation (DIC) data, we extract essential features representing crack initiation and propagation dynamics. These features are then integrated into a physics-informed neural network (PINN) trained to predict fatigue life remaining (FLR) and crack length evolution over time. The approach demonstrates superior performance compared to traditional empirical models and finite element simulations, offering a faster and more accurate prediction capability crucial for structural integrity assessment and design optimization.